Work / independently operated systems
Every project here is a deployed system or research artifact. The through-line is simple: make high-consequence technology safer, more observable, and easier to operate.
01 / Featured case studies
A private cloud built entirely as code — Proxmox, Terraform, Traefik, and 50+ services with secrets management and observability.
5 VMs, a Raspberry Pi, and 53 Docker/Kubernetes services behind Traefik with Let's Encrypt. Entire platform is Terraform-managed; OpenBao (Vault fork) handles secrets with automated unseal; real-time dashboards and uptime alerting across everything.
A production incident-notification system: FastAPI + SQLite dashboard, n8n routing, ntfy push, six systemd-timer producers, and encrypted backups.
Full lifecycle project — designed the event model, fixed a real concurrency bug (asyncio lock + SQLite pool exhaustion), shipped six timer-driven producers (service health, Proxmox, storage, backup, Grafana, Pi), and a daily GPG-encrypted backup with a validated restore drill.
A persistent AI agent (DeepSeek backend, Telegram interface, memory) extended with a custom skill that manages tasks across Obsidian vaults.
Runs 24/7 with persistent memory and MCP integrations (JobSync, PriceBuddy). The TaskNotes skill teaches the agent to create and manage tasks across 15 vaults via a mounted filesystem, with context-aware @Work/@Personal/@Homelab tagging.
02 / Supporting systems
Prometheus + Grafana + Blackbox across the platform.
Unified metrics for VMs, containers, Proxmox, and every service, with blackbox uptime/latency probes and alerting.
JobSync, PriceBuddy, jobspy-mcp, arr-controlarr — MCP as the agent tool-access layer.
Two MCP servers built and eight more operated as the access layer for agents (Hermes, OpenCode), with bearer-token auth and Traefik-only exposure.
HA bare-metal cluster across two Nutanix nodes running Proxmox.
Simulates enterprise HPC environments for HPC/LLM workloads, bridging bare-metal hardware and cloud-first platform engineering.
Local LLM inference, fine-tuning, and RAG deep research.
OpenAI-compatible local serving, parameter-efficient fine-tuning of open-weight models, and RAG-based deep research with cited synthesis.
OpenBao for secrets, BorgBackup to off-site storage.
Centralized secret management with auto-unseal, plus encrypted off-site backups for Nextcloud and notification-state.